The semiconductor industry has never experienced such a prosperous period! Artificial intelligence has created an insatiable demand for chips used to train and run artificial intelligence models. In the first half of 2026, annual semiconductor sales growth exceeded any period since the 1980s. The British "Economist" reported that almost all industries are expected to grow dependent on artificial intelligence chips, which is undoubtedly a huge bet on this technology. If this bet fails, as some call the end of the "supercycle," will chipmakers suffer a downturn as sharp as the boom? This media analysis points out that there are reasons to think this will not happen.
TSMC, the world's largest chip contract manufacturer, announced record profits for the sixth consecutive quarter; South Korean chip giant Samsung Electronics also achieved record profits for the fourth consecutive quarter. However, investors are increasingly uneasy. Despite the impressive results and optimistic outlook, the index of major semiconductor companies has fallen by more than 10% since October 6.
For more than 40 years, the semiconductor industry has been guided by two laws as reliable as the laws of physics. The first is that the number of transistors (tiny switches etched onto wafers) continues to grow. The second is that booms will eventually end in busts. The first rule is that chip performance improvements are currently in a booming stage. Leading artificial intelligence processors contain more than 200 billion transistors, and industry insiders estimate that the number of transistors could reach 1 trillion by the end of this decade. The second law raises concerns about the sustainability of this boom.
Although the semiconductor industry is booming, the concentration of the manufacturing sector has increased significantly. Most chip design companies, such as Huida, which supplies graphics processors for the artificial intelligence boom, do not make their own chips. The same goes for software companies such as Google and OpenAI, which have also begun designing artificial intelligence processors. Instead, they rely on foundries. At the beginning of the 21st century, there were approximately 25 companies capable of producing state-of-the-art processors. Today, only four companies can do this, and TSMC almost handles the production of all high-end processors.
Memory chips are also critical to artificial intelligence, but their development trajectories are also similar. In 1990, more than 20 companies produced advanced memory chips. Today, only three companies, Samsung, SK Hynix and Micron, almost monopolize the market. China has new companies emerging in both processor and memory chip fields, but they still lag behind the industry's forefront due to a lack of state-of-the-art chip manufacturing equipment.
Industry consolidation has allowed a handful of dominant chipmakers to control the pace of expansion and dictate the terms of sales. They are investing heavily to capitalize on the opportunities presented by this boom. TSMC’s capital expenditures this year are expected to reach approximately US$62 billion, more than 1.5 times that in 2025. Analysts expect that number to reach about $76 billion next year. Memory makers have also stepped up investment.
Even so, their expenditures are still quite limited compared with the huge benefits brought by artificial intelligence. TSMC’s capital expenditures as a percentage of revenue are expected to fall from 44% in 2023 to 36% this year.
In addition, these investments are paid for with cash flow generated by the chip manufacturers themselves, rather than through borrowing. TSMC's capacity expansion will consume a little more than half of its operating cash flow this year, while Samsung's ratio will be just over a fifth. The chipmakers' debt pales in comparison to their growing cash hoards.
Bill Wiseman of consulting firm McKinsey said manufacturers are enjoying a very favorable price environment and have every reason to control the pace of expansion to maintain that advantage. DRAM prices are approximately six times what they will be in early 2023. TSMC has not lowered its prices since launching its three-nanometer process in late 2022, although typically the cost of a process drops after a few years.
Wiseman noted that the pace of capacity expansion has also been hampered because many new chip fabs are being built in the United States, where construction cycles are "about twice as long as elsewhere."
Industrial consolidation also gives chip manufacturers greater bargaining power, changing who bears the losses caused by the economic slowdown. Until recently, memory was sold primarily through the spot market or short-term contracts. When demand is weak, customers are under no obligation to keep buying, so prices plummet.
Today, with capacity scarce and suppliers few and far between, buyers need to sign years in advance and accept the seller's terms. Huida has committed more than US$250 billion to ensure supply and parts availability through January 2029. Micron Technology said in September that a "take-or-pay" agreement, in which customers must buy an agreed-upon quantity or otherwise pay, has locked in 35% of its revenue until 2030, with commitments reaching $32 billion, most of which involves cash advances. Such advances help finance new plants. They also transfer at least some of the risk of a market crash to buyers.
The end of the artificial intelligence boom will still cause a lot of pain for chipmakers. If spending stalls, prices will fall, profits will shrink, and stock prices that were once ridiculously high will return to rationality. Clients will seek to renegotiate contracts signed at market peaks. The second law of the semiconductor industry still applies. This time, however, customers will share in the consequences of the market crash.
